The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes shell commands for environment setup, building, linting, and testing. These commands are sourced dynamically from a .claude/routines.md file within the target repository.
[INDIRECT_PROMPT_INJECTION]: The skill processes a repository-local configuration file (.claude/routines.md) to determine which commands to execute. This creates an attack surface where a malicious repository could provide harmful commands to be executed by the agent.
Ingestion points: .claude/routines.md is loaded and its commands are executed during the preflight and verification stages (SKILL.md).
Boundary markers: None identified; the skill assumes the configuration file is authoritative and safe.
Capability inventory: The skill has the ability to execute arbitrary shell commands via bash, modify repository files via git, and interact with forge APIs (GitHub/GitLab) via gh or glab CLI tools.
Sanitization: There is no evidence of command sanitization or validation against a whitelist.
[SAFE]: The skill uses well-known forge tools (gh and glab) for authentication and pull request creation. These tools target trusted platforms (GitHub and GitLab).
[SAFE]: The skill implements a robust safety invariant: it is strictly forbidden from merging code automatically, ensuring all automated changes require human approval via a pull request review process.

maintenance-routine